METHOD AND DEVICE FOR PROVIDING MEDICAL INFORMATION

Description

[0001] The invention relates to methods and devices for providing medical information. The medical information may, for example, include information on possible clinical findings or other clinically relevant information for a patient. The medical information may be provided based on a (semi-)automated analysis of the medical data sets available to a user.

[0002] Treating a patient involves numerous medical decisions that directly impact the patient's well-being. The decisions to be made are complex. They include, among other things, treatment decisions and prognostication. Depending on the case, follow-up examinations or referrals may also need to be decided. For all of these decisions, a diagnosis of the patient that is as accurate as possible, based on the available information, is essential.

[0003] In the modern hospital environment, medical staff have virtually unlimited access to all types of information related to a given case. As medical information systems become increasingly widespread and powerful, potentially decision-relevant data can, in principle, be accessed at any time. This includes, for example, access to medical guidelines, databases of similar cases, or the electronic patient record with all previous examinations and findings. This has the disadvantage that the volume of information has become so extensive that it is almost impossible for medical staff to keep track of it. This is where so-called "clinical decision support systems" come into play.Based on structured data storage and concerted data retrieval, such systems are designed to filter out relevant information, evaluate it and make it available to the user, for example in the form of suggestions.

[0004] US 2020 / 0 380 675 A1 discloses an automated pipeline for the accurate detection and segmentation of lesions. US 2016 / 0 267 221 A1 discloses methods comprising retrieving medical reference images for a medical image of a patient to support diagnosis. US 2012 / 0 114 256 A1 discloses a system for retrieving comparison cases based on a similarity analysis of medical image data. US 2019 / 0 117 978 A1 discloses a system for retrieving clinical information based on a similarity analysis of electrocardiograms.

[0005] One problem with existing systems is that they can only work with data to which they have access. If decision-relevant data is missing, the information or suggestions provided to the user may be incomplete or, in the worst case, incorrect. This can be particularly the case when, based on the available data, many different conclusions are equally possible. Taking the example of deriving one or more possible medical diagnoses, this can result in a comparatively large number of possibilities that, considered individually, have no significant relevance to the case at hand.A meaningful ranking of the medical diagnoses in question is often not possible, since small differences in relevance tip the scales in favor of a possible diagnosis that, objectively speaking, does not have a significantly higher probability of being correct than other possible diagnoses.

[0006] Against this background, one object of the invention is to provide improved methods and associated devices capable of automatically deriving medical information about a patient from the available medical data and providing it to a user. In particular, the methods and devices should be able to provide medical information with higher confidence and less uncertainty.

[0007] According to the invention, the stated object is achieved with a method, a device, a computer program product, or a computer-readable storage medium according to the main claim and the independent claims. Advantageous developments are specified in the dependent claims. The inventive solution to the object is described below both with reference to the claimed devices and with reference to the claimed methods. Features, advantages, or alternative embodiments / aspects mentioned here are also to be applied to the other claimed subject matter, and vice versa. In other words, the subject claims (which are directed, for example, to a device) can also be developed with the features described or claimed in connection with a method. The corresponding functional features of the method are implemented by corresponding subject modules.

[0008] Furthermore, the inventive solution to the problem is also described with reference to methods and devices for adapting trained functions. In this case, features and alternative embodiments / aspects of data structures and / or functions in methods and devices for determination can be transferred to analogous data structures and / or functions in methods and devices for adaptation. Analogous data structures can be identified in particular by the use of the prefix "training." Furthermore, the trained functions used in methods and devices for providing the medical information can be adapted and / or provided in particular by methods and devices for adapting trained functions.

[0009] According to one aspect, a computer-implemented method for providing medical information is provided. The method comprises several steps. A first step is directed towards receiving a medical data set of a patient. A further step is directed towards determining one or more reference data sets from a plurality of comparison data sets based on similarity measures. A similarity measure is based on the similarity of the data set and one of the comparison data sets. In addition, each comparison data set is associated with at least one previously known medical finding. A further step is directed towards identifying one or more medical attributes, which medical attributes each comprise a characteristic of one or more of the previously known medical findings associated with the reference data sets.A further step is directed toward displaying the identified medical attributes via a user interface. A further step is directed toward receiving user input from a user regarding the displayed medical attributes via the user interface. A further step is directed toward generating medical information for the patient based on the medical findings associated with the reference data sets and the user input. A further step is directed toward providing the medical information.

[0010] The user input comprises a verification of at least some of the identified medical attributes, comprising a confirmation and / or rejection of one or more of the identified attributes.

[0011] The medical data set can be provided or selected by a user. The user can, for example, be a physician, such as a radiologist, who wants to make a medical diagnosis for the patient (the patient is also referred to below as the "patient to be diagnosed"). Furthermore, the user can also be the patient themselves. The medical data set can be received by the user, for example, by uploading the medical data set. Alternatively, the medical data set can be received from one or more corresponding databases in which the entire medical data set or parts of it are stored.The databases can, for example, be part of medical information systems, such as hospital information systems (the English technical term for this is hospital information system or HIS for short) and / or PACS systems (PACS stands for picture archiving and communication system) and / or laboratory information systems (LIS).

[0012] The medical dataset contains medical data available for the patient. The medical data can include both medical image data and non-image data. Medical image data is image data acquired using an imaging modality and can, in particular, depict a body part of the patient. Imaging modalities can include, for example, computed tomography devices, magnetic resonance imaging devices, X-ray devices, ultrasound devices, and the like. Furthermore, medical image data can include digitized histopathology images depicting a suitably prepared tissue section of the patient. The image data can further include longitudinal data, for example in the form of time series or temporally spaced follow-up images. Non-image data can include, in particular longitudinal, data containing one or more medical values of the patient and / or elements from the patient's medical history.This may include laboratory data, vital signs, and / or other patient-related measurements or previous examinations. Furthermore, non-image data may include demographic information related to the patient, such as age, gender, lifestyle habits, risk factors, etc. Furthermore, non-image data may include one or more previous medical findings and / or other assessments (e.g., from other, possibly referring physicians). These may be included in the medical dataset, for example, in the form of one or more structured or unstructured medical findings.

[0013] According to some embodiments / aspects of the invention, the medical dataset may contain very comprehensive information about the patient's health status (but according to other embodiments / aspects, it may also be limited to only one data category—such as image data, particularly radiology data). The user's task may be to create a medical finding, diagnosis, or conclusion based on the medical dataset.

[0014] To support the user, the initial plan is to use automatic processing of the medical data set to search for similar cases for which the findings / diagnoses / conclusions already made are (pre-)known. This is based on the idea that findings from similar cases can potentially be relevant for the current case. To this end, the plan is to identify reference data sets from a set of comparison data sets which reference data sets show a certain similarity to the medical data set. The comparison data sets can have a similar structure to the medical data set, i.e. in particular they can include image data and non-image data. For each of the comparison data sets, at least one medical finding is pre-known. The pre-known medical findings can, for example, have been verified or annotated by one or more physicians.In particular, they may have been verified by the user themselves in the past. The comparison data sets may be associated with any medical findings. Alternatively, the medical findings may be selected from a predefined set of medical findings. The comparison data sets may be stored in one or more databases, which may also be part of medical information systems. In particular, the comparison data sets may correspond to real patient data, which has preferably been anonymized so that the underlying patient can no longer be identified. Alternatively or additionally, the comparison data sets may be synthetic data sets that are not based on real patients. Furthermore, the comparison data sets may be derived from medical guidelines, compendia, or textbooks in which medical findings and their clinical signs are prototypically specified.

[0015] To determine the reference datasets, all available comparison datasets can be examined for their similarity to the medical dataset. Alternatively, a pre-selection of the comparison datasets can be made based on the medical dataset, whereby only the pre-selected comparison datasets are considered when identifying the reference datasets. If the medical dataset contains histopathology image data, for example, this can pre-select comparison datasets that also contain histopathology image data, which can make processing more efficient and shorten the response time. For each of the considered comparison datasets, a similarity measure can be determined. This similarity measure is based on a similarity between the medical dataset and the respective comparison dataset and, in particular, indicates or quantifies a similarity.A similarity measure can, for example, be a numerical value or "score." Similarity measures can, for example, be determined based on the application of a similarity metric that outputs a similarity measure based on the input variables, i.e., the medical data set and a comparison data set. The similarity metric can, in particular, be implemented in a (first) data processing algorithm. Reference data sets are, in particular, those comparison data sets that exhibit a certain similarity to the medical data set. In other words, reference data sets can, in particular, be those comparison data sets whose similarity measure lies above a predetermined, predefined, or specifiable threshold.

[0016] Since each comparison data set is associated with at least one medical finding, the automatic search for similar data sets provides a selection of medical findings that may be relevant for the patient being diagnosed. Furthermore, the similarity measures can be used to estimate which medical findings from this selection are potentially more relevant than others. This is based on the idea that high similarity measures indicate a high relevance of the respective comparison data set for the present case. Therefore, optionally, based on the similarity measures, a relevance value can be calculated for one or more of the previously known medical findings associated with the reference data sets. This relevance value is based in particular on the relative relevance of the respective medical finding compared to other medical findings, or indicates and / or quantifies such a relevance.

[0017] To further narrow down the possible medical findings, the user is to be involved in a continuous and guided human-machine interaction. To this end, one or more medical attributes are to be identified for each possible medical finding (i.e., the medical findings associated with the reference data sets), which can then be easily checked by the user for their application to the current case. In particular, several different medical attributes can be identified for each medical finding. The medical attributes can describe higher-level properties of the medical finding. In other words, the medical findings can be understood as characteristics, particularly clinical ones, of the respective medical finding. Several different medical attributes can be assigned to a medical finding.In summary, the medical attributes can characterize the medical finding assigned to them and, in particular, differentiate it from other medical findings / differential diagnoses. In particular, the medical attributes can be selected so that the user can easily decide whether the medical attribute is relevant to the current case. The medical attributes can be divided into one or more categories. In particular, the medical attributes can be determined and assigned such that the combination of medical attributes assigned to a medical finding makes it uniquely identifiable. The identified medical attributes can be arbitrary and, in particular, can be determined dynamically depending on the medical findings associated with the reference data sets.Alternatively, the medical attributes can be selected from a predetermined set of medical attributes in the identification step. The medical attributes can be identified based on a predefined and deterministic rule. The medical attributes can be identified by a data processing algorithm that assigns one or more medical attributes to each of the medical findings associated with the reference data sets. The data processing algorithm can be implemented, for example, as an electronic classifier.

[0018] To give the user the opportunity to check whether or which of the identified medical attributes are applicable to the present case, and to what extent they are applicable, the identified medical attributes are displayed to the user via a user interface. The user interface can be designed accordingly, for example, by implementing a graphical user interface with corresponding display functions. Furthermore, the user can enter user input regarding the displayed medical attributes via the user interface. The user interface can also be designed accordingly for this purpose, for example, by implementing a graphical user interface with corresponding input functions.

[0019] Since the user input relates to the medical attributes, which in turn are characteristics of the medical findings associated with the reference data sets, the user input has a potential impact on the relevance of the medical findings associated with the reference data sets for the patient to be diagnosed. This is exploited to generate medical information based on the medical findings. The medical information can, in particular, be information relating to the patient's diagnosis and, for example, indicate whether one or more of the medical findings are relevant for the patient to be diagnosed or which medical findings can be excluded. The medical information can, in particular, indicate which of the medical findings the user input indicates.Furthermore, the medical information may contain additional information extracted from the medical data set and / or the reference data sets. This additional information may, for example, be based on those data objects in the medical data set and the reference data sets that are identified as similar.

[0020] The step of providing makes the medical information available for further use. For example, the medical information can be provided for presentation of the medical information to the user via the user interface. Alternatively or additionally, the medical information can be provided for storage in a storage device or for further data processing. In particular, the medical information can be provided for improving a data processing algorithm for selecting the reference data sets or a data processing algorithm for identifying the medical attributes. Furthermore, based on the medical information, the user can make further user inputs directed at the medical attributes, whereupon new medical information can be determined or existing medical information can be adapted.

[0021] The features work synergistically to support the user in making a medical diagnosis. In other words, a method for making a medical diagnosis is provided by an automated system that processes physiological measurements (in the form of the medical data set and the comparison data sets). A two-stage approach is followed. First, possible medical findings are narrowed down through a similarity analysis of the available data. This is the prerequisite for supporting the user in the subsequent decision-making process as to the extent to which the possible medical findings are relevant for the patient being diagnosed, particularly through iterative human-machine interaction.The decision-making process is systematically broken down into sub-problems that are easier for the user to understand through the automatic identification of medical attributes. This is because it is typically easier for the user to decide whether the higher-level medical attributes apply to the specific case than it is for a complete medical finding. The medical attributes are objective characteristics of the respective medical finding and are therefore independent of subjective user perceptions. Nevertheless, the result is not deterministic or predetermined, but depends on the specific user input. In this way, an improved statement can be made about which medical findings are relevant for the patient being diagnosed, thereby improving both the confidence and the accuracy of the medical information provided.

[0022] In other words, a method and corresponding device for providing medical information are disclosed. Based on the available data on a patient to be diagnosed, possible medical findings are first determined. Medical attributes are then identified for the possible medical findings, which characterize the possible medical findings in the context of the patient to be diagnosed. The medical attributes are then displayed to a user for evaluation / verification. A corresponding user input is used to evaluate the relevance of the possible medical findings.

[0023] According to another aspect, a computer-implemented method for providing medical information is provided, comprising several steps as follows: Receiving a patient's medical data set; accessing multiple comparison data sets, each of which is associated with a previously known medical finding; calculating a similarity measure for each comparison data set, each similarity measure indicating a similarity between the respective comparison data set and the medical data set; identifying one or more medical attributes based on the similarity measures, each of which medical attributes comprises a characteristic of one or more of the previously known medical findings; displaying the identified medical attributes via a user interface; receiving user input from the user regarding the displayed medical attributes via the user interface; generating medical information for the patient based on the previously known medical findings as well as the user input and / or the similarity measures;and providing medical information.;

[0024] Generating the medical information includes determining a ranking of at least a portion of the previously known medical findings associated with the reference data sets based on the similarities determined for the respective reference data sets and / or based on the user input.

[0025] The ranking provides a statement about the relative relevance of the medical findings for the patient being diagnosed. This allows a differentiated assessment of the processing results (both by the user and by subsequent processing steps), which can improve both the confidence and accuracy of the provided medical information. The ranking can include calculating a relevance or confidence value based on the similarity measures for each medical finding and / or the user input. Either only the user input or only the similarity measures can be taken into account. Furthermore, a synthesis of these two pieces of information can occur, for example by adjusting relevance or confidence values based on the similarity measures using the user input.After each iteration of the procedures, in particular after each user input directed at the identified attributes, the ranking can be re-determined.

[0026] According to one aspect, providing the medical information comprises displaying the medical information via the user interface, wherein the display of the medical information is based on the ranking and in particular in a form of representation that is designed such that the user can at least partially perceive the ranking of the medical findings.

[0027] This has the technical effect that the relative relevance of the respective medical finding is communicated to the user during the ongoing and guided human-machine interaction and can thus be taken into account in further analysis. The user is supported in the technical task of making a medical diagnosis based on the analysis of medical data. The support provided by the feature is causally linked to the objective result derived from the processing, the medical information. The display of this information as such does not depend on subjective preferences. There are various ways in which the display can be designed based on the ranking and, in particular, the form of presentation. For example, the medical findings can be displayed in the form of a list with descending or ascending relative relevance.In addition, it is possible to number the medical findings according to their position in the ranking and display the ranking number. Furthermore, the display can include displaying the aforementioned confidence or relevance value. This has the additional technical effect of providing the user with information about how the relative relevance of the medical findings differs from one another. In principle, all or only some (preferably the most relevant) medical findings can be displayed in the display step.

[0028] According to one aspect, the methods further comprise a step of providing additional information about one or more of the medical findings, and displaying the medical information includes displaying the additional information.

[0029] The additional information can include higher-level information or meta-information on the respective medical finding, which can be provided, for example, from electronic textbooks, guidelines, or compendia. The additional information can, for example, include information on the distribution of medical findings in different cohorts and / or information on further procedures, such as treatment options or additional examination steps. This automatically provides the user with additional information that can, on the one hand, make the result of the analysis more transparent and, on the other hand, suggest further steps. This provides the user with targeted support in making a medical diagnosis. In particular, since the user does not have to obtain this information themselves, the process can be more efficient and standardized.

[0030] According to one aspect, the additional information can be displayed in such a way that the respective additional information can be assigned by the user to the respective medical finding.

[0031] According to one aspect, the methods further comprise a step of extracting one or more data objects from one or more reference data sets and / or the medical data set, and displaying the medical information includes displaying the data objects. It is preferred that these are, in particular, data objects on the basis of which the similarity measures were determined. In other words, the data objects can relate to the similarities between the reference data sets and the medical data set. For example, the data objects can comprise image data depicting similar pathological changes (in particular, similar pathological changes as in the image data of the medical data set). The image data can be displayed in the display step, for example, as (pixel) images.

[0032] By extracting and displaying the data objects, the user can better assess the quality of the automatic similarity analysis and, in particular, base user input on a broad information base. This improves human-machine interaction.

[0033] According to one aspect, the data objects can be displayed in such a way that the user can assign each data object to the respective medical finding. In particular, the data objects can be grouped according to the medical findings.

[0034] In one aspect, the medical attributes include one or more of the following: an indication, in particular demographic information, regarding a patient group associated with the clinical finding; an indication regarding one or more symptoms associated with the respective medical finding; an indication regarding one or more differential diagnoses associated with the respective medical finding; an indication regarding one or more clinical signs associated with the respective medical finding; and / or an indication regarding one or more clinical contraindications associated with the respective medical finding.

[0035] The aforementioned information can also be referred to as categories of medical attributes. A demographic statement can, for example, be an age group in which a medical finding typically occurs. Furthermore, a demographic statement can include gender or typical lifestyle habits and / or risk factors of the patient group typically associated with the medical finding and can indicate, for example, whether the patients are smokers. The clinical signs or countersigns can, among other things, relate to the course of a disease and indicate, for example, whether a medical finding is associated with an acute, subacute, or chronic illness. Furthermore, clinical signs or countersigns can relate to the location of a pathological change and indicate, for example, whether lesions in the lung occur unilaterally, bilaterally, or focally.Furthermore, clinical signs or contraindications can refer to the frequency of occurrence of a medical diagnosis. Such medical attributes allow medical findings to be characterized multilaterally, and user input directed at the medical findings allows for a differentiated assessment, which can improve both the confidence and accuracy of the provided medical information.

[0036] According to one aspect, the methods comprise, in the step of identifying the one or more medical attributes, providing an assignment that assigns one or more medical attributes to at least one previously known medical finding, and assigning one or more medical attributes to at least one of the medical findings associated with the reference data sets using the assignment. In particular, in the step of assigning, several different medical attributes can be assigned to each medical finding associated with the reference data sets.

[0037] The assignment can, in particular, be predefined. For example, the assignment can be in the form of a list that assigns one or more, and in particular several different, medical attributes to each medical finding. The list can, for example, be stored in a storage unit. Furthermore, the assignment can be designed such that the medical attributes are stored in the respective comparison data set according to the respectively associated medical findings. By providing the assignment, rapid and objective identification of the medical attributes can be achieved, thereby improving the confidence and accuracy of the medical information provided.

[0038] The step of identifying the one or more medical attributes comprises determining an attribute ranking based on the similarity measures determined for the respective reference data sets and / or based on a discriminatory effect of the medical attributes with regard to the relevance of the medical findings for the patient to be diagnosed and / or based on one or more different attribute categories, wherein the display of the identified medical attributes takes place in a form of representation that is designed such that the user can perceive the attribute ranking.

[0039] This provides the user with additional information about how crucial a medical attribute is for the medical information. This makes the analysis results more understandable for the user. It also allows the user to focus on medical attributes and, if necessary, prioritize their inputs, which can make human-machine interaction more efficient. The discriminatory effect can be determined, for example, by analyzing possible effects on the ranking or simulating possible decision trees that lead to the confirmation of one of the medical findings.

[0040] According to one aspect, the display of the identified medical attributes may be performed in such a way that the medical attributes are grouped by categories, which provides the user with a better overview of the medical attributes.

[0041] In principle, it can include all identified medical attributes or only a part of the identified medical attributes, whereby the identification step can already include a selection from the medical attributes assigned to the medical findings.

[0042] According to one aspect, the methods further comprise a step of generating incident medical information based on the medical findings associated with the reference data sets, in particular based on the similarities determined for the respective reference data sets, wherein the step of generating the medical information comprises adapting the incident medical information based on the user input.

[0043] Incident medical information can be generated essentially as described above in connection with medical information and, in particular, can contain the same or similar components. Incident medical information differs from medical information in that it does not take user input into account. In other words, it is based solely on similarity analysis. By adapting it based on user input, medical information can then be created, thereby improving the confidence and accuracy of the provided medical information in a multi-step process.

[0044] According to one aspect, the methods further comprise a step of displaying the incident medical information, in particular together with the identified medical attributes.

[0045] This provides the user with information about the results of the similarity analysis and allows them to better assess the basis on which the medical attributes were selected. This allows them to base their user input on a broader information base. This can make the procedures more user-friendly and efficient. Furthermore, it can increase the user's confidence in the analysis.

[0046] According to one aspect, the user input includes a verification of at least some of the identified medical attributes and, in particular, an evaluation of one or more of the identified attributes. According to further aspects, the user input further includes a weighting of one or more of the identified medical attributes.

[0047] By selecting or deselecting individual medical attributes, the user can decide which of the medical attributes apply to the specific case. A rating also allows for gradual grading. This results in a pattern of medical attributes for the specific case, which can be compared with the attribute pattern of the medical findings in order to identify relevant medical findings and exclude less relevant ones. By weighting, the user can, for example, additionally indicate how crucial they consider the respective medical attribute to be for the decision-making process. Medical attributes that are important to the user can, for example, be given a higher weighting than others when generating or adapting the medical information (e.g. twice).In other words, the step of generating medical information is then also based on weighting. A user input regarding a medical attribute with a high weighting can have a greater influence on the medical information. This allows the user input to be evaluated even more specifically, thereby improving the confidence and accuracy of the provided medical information.

[0048] According to one aspect, the medical data set and the comparison data sets each comprise medical image data, and a similarity measure is each based on a similarity between the medical image data of the medical data set and the medical image data of a comparison data set.

[0049] In other words, similarity analysis aims to find similar image data to the image information in the available medical dataset. Image information is compared to identify similar patients. This has been shown to be a fast and accurate approach to identifying reference datasets and thus inferring potentially relevant medical findings, ultimately improving the confidence and accuracy of the medical information provided.

[0050] According to one aspect, determining the one or more reference data sets comprises the steps of: Extracting a data descriptor from the medical data set; receiving a corresponding data descriptor for each of the comparison data sets; determining, for each comparison data set, a similarity measure, wherein a similarity measure is based on a similarity between the data descriptor and a corresponding data descriptor; and determining the one or more reference data sets based on the determined similarity measures.

[0051] The data descriptor can comprise one or more features extracted from or calculated from the medical dataset, and in particular from the image data of the medical dataset. In addition, the data descriptor can be based on (or additionally taking into account) further information or non-image data, such as metadata relating to the image data, patient data, other measured values, medical findings, etc. Another term for data descriptor can be the term "feature signature." The data descriptor can, in particular, characterize the medical dataset. The features of the data descriptor can be combined into a feature vector. In particular, the data descriptor can comprise such a feature vector. Features extracted from image data can be morphological and / or structural and / or texture-related and / or pattern-related features.Features extracted from non-image data may be related to a finding, a medical report, a measured value, demographic information, etc. The (first) data processing algorithm may, in particular, be configured to determine similarity measures based on the data descriptor.

[0052] Determining the similarity measures can comprise extracting or receiving a corresponding data descriptor from each of the possible comparison data sets. The procedure can be the same as for the data descriptor extracted based on the medical data set. Furthermore, determining the similarity measures can comprise comparing the corresponding data descriptors with the respective data descriptor. The comparing step can be based, in particular, on determining a distance between the respective data descriptors, calculating a cosine similarity of the data descriptors, and / or calculating a weighted sum of the difference or similarity of individual features of the data descriptor. In particular, those comparison data sets whose associated similarity measure is greater than a predefined or predefinable threshold can be identified as reference data sets.

[0053] By using data descriptors, easy-to-implement and easily transferable parameters are defined for comparing different data sets. Furthermore, the features contained in the feature signatures can be based on higher-level observables derived from the data sets, which often characterize the properties of the data sets better than the underlying data itself.

[0054] According to one aspect, determining the one or more reference data sets comprises applying a trained function, which trained function is configured to determine a similarity measure between medical data sets.

[0055] A trained function generally maps input data to output data. In this case, the output data can in particular depend on one or more parameters of the trained function. The one or more parameters of the trained function can be determined and / or adapted through training. The determination and / or adaptation of the one or more parameters of the trained function can in particular be based on a pair of training input data and associated training output data, wherein the trained function is applied to the training input data to generate training mapping data. In particular, the determination and / or adaptation can be based on a comparison of the training mapping data and the training output data. In general, a trainable function, i.e. a function with parameters that have not yet been adapted, is also referred to as a trained function.In particular, the trained function can be included in the (first) data processing algorithm.

[0056] Other terms for trained functions are trained mapping rule, mapping rule with trained parameters, function with trained parameters, algorithm based on artificial intelligence, and machine learning algorithm. An example of a trained function is an artificial neural network. Instead of the term "neural network," the term "neural net" can also be used. A neural network is essentially structured like a biological neural network—such as a human brain. In particular, an artificial neural network comprises an input layer and an output layer. It can also comprise multiple layers between the input and output layers. Each layer comprises at least one, preferably several, nodes. Each node can be understood as a biological processing unit, e.g., a neuron.In other words, each neuron corresponds to an operation applied to input data. Nodes in one layer can be connected to nodes in other layers by edges or connections, particularly by directed edges or connections. These edges or connections define the data flow between the nodes of the network. The edges or connections are associated with a parameter, often referred to as a "weight" or "edge weight." This parameter can regulate the importance of the output of a first node for the input of a second node, where the first node and the second node are connected by an edge. In particular, a trained function can also comprise a deep artificial neural network (the technical term is "deep neural network").

[0057] In particular, a neural network can be trained. Specifically, the training of a neural network is performed based on the training input data and the corresponding training output data according to a "supervised" learning technique (the technical term is "supervised learning"), whereby the known training input data is fed into the neural network and the output data generated by the network is compared with the corresponding training output data. The artificial neural network learns and adjusts the edge weights for each node independently as long as the output data of the last network layer does not sufficiently match the training output data.

[0058] According to one aspect, at least one of the trained functions comprises a convolutional neural network, and in particular a region-based convolutional neural network.

[0059] The technical term for a convolutional neural network is a convolutional neural network. In particular, a convolutional neural network can be designed as a deep convolutional neural network. The neural network has one or more convolutional layers and one or more deconvolutional layers. In particular, the neural network can include a pooling layer. By using convolutional layers and / or deconvolutional layers, a neural network can be used particularly efficiently for image processing, since, despite the many connections between node layers, only a few edge weights (namely, the edge weights corresponding to the values of the convolution kernel) need to be determined.With the same amount of training data, the accuracy of the neural network can also be improved.

[0060] A region-based convolutional neural network is a so-called region-based convolutional neural network. A region-based convolutional neural network can be a so-called fast region-based convolutional neural network or a faster region-based convolutional neural network. Region-based convolutional neural networks are characterized by having integrated functionalities for defining potentially relevant data regions, making them suitable for section-by-section determination of similarities according to embodiments / aspects of the invention.

[0061] According to one aspect, the step of providing comprises providing the medical information to the trained function and further comprising the step of adapting the trained function based on the medical information.

[0062] In particular, medical information includes user input regarding medical attributes. Since a similar pattern of medical attributes indicates similar medical data sets, the medical information can be used to continuously improve the trained function during use (a term used in English is 'continuous learning').

[0063] According to one aspect, a system for providing medical information is provided. The system comprises an interface and a controller. The controller is configured to receive a medical data set of a patient via the interface. The controller is further configured to determine one or more reference data sets from a plurality of comparison data sets based on similarity measures, wherein a similarity measure is based on a similarity between the medical data set and a comparison data set, and wherein each comparison data set is associated with at least one previously known medical finding. The controller is further configured to identify one or more medical attributes, which medical attributes each comprise a characteristic of one or more of the previously known medical findings associated with the reference data sets.The controller is further configured to display the identified medical attributes (via the interface). The controller is further configured to receive user input from a user regarding the displayed medical attributes via the interface. The controller is further configured to generate medical information for the patient based on the medical findings associated with the reference data sets and the user input. The controller is further configured to provide the medical information via the interface.

[0064] The controller can be configured as a centralized or decentralized processing unit. The processing unit can have one or more processors. The processors can be configured as a central processing unit (CPU) and / or a graphics processing unit (GPU). Alternatively, the controller can be implemented as a local or cloud-based processing server. Furthermore, the controller can include one or more virtual machines.

[0065] The interface can generally be designed for data exchange between the controller and other components. The interface can be implemented in the form of one or more individual data interfaces, which can have a hardware and / or software interface, e.g. a PCI bus, a USB interface, a FireWire interface, a ZigBee or a Bluetooth interface. The interface can furthermore have an interface of a communications network, wherein the communications network can have a local area network (LAN), for example an intranet or a wide area network (WAN). Accordingly, the one or more data interfaces can have a LAN interface or a wireless LAN interface (WLAN or Wi-Fi). The interface can furthermore be designed for communication with the user via a user interface.Accordingly, the controller may be configured to display the medical attributes via the user interface and to receive the user input via the user interface.

[0066] The advantages of the proposed device essentially correspond to the advantages of the proposed method. Features, advantages, or alternative embodiments / aspects can also be transferred to the other claimed subject matter, and vice versa.

[0067] According to one aspect, the system further comprises a database for storing a plurality of comparison data sets. The interface is in data communication with the database. The controller is further configured to access the comparison data sets to determine the reference data sets.

[0068] The controller is further configured to select comparison data sets from the database based on the medical data set.

[0069] The database can be configured as a centralized or decentralized storage unit. The database can, in particular, be part of a server system. The database can, in particular, be part of a medical information system, such as a hospital information system (HIS) and / or a PACS system (PACS stands for picture archiving and communication system) and / or a laboratory information system (LIS). The database can also be configured as a cloud storage system.

[0070] According to a further aspect, a diagnostic system is provided, which comprises the system for providing medical information and a medical information system configured to store and / or provide medical data (in particular the medical data set and the comparison data sets). The medical information is connected to the system for providing medical information via the interface. Furthermore, the medical information system can comprise one or more imaging modalities, such as a computed tomography system, a magnetic resonance system, an angiography system, an X-ray system, a positron emission tomography system, a mammography system, and / or a system for generating histopathology image data.

[0071] In a further aspect, the invention relates to a computer program product which comprises a program and is loadable directly into a memory of a programmable controller and has program means, e.g. libraries and auxiliary functions, in order to carry out a method for providing similarity information, in particular according to the aforementioned embodiments / aspects, when the computer program product is executed.

[0072] Furthermore, in a further aspect, the invention relates to a computer-readable storage medium on which readable and executable program sections are stored in order to carry out all steps of a method for providing similarity information according to the aforementioned embodiments / aspects when the program sections are executed by the controller.

[0073] The computer program products can comprise software with source code that still needs to be compiled and linked, or that only needs to be interpreted, or executable software code that only needs to be loaded into the processing unit for execution. The computer program products enable the methods to be executed quickly, identically repeatably, and robustly. The computer program products are configured such that they can execute the method steps according to the invention using the computing unit. The computing unit must have the prerequisites, such as appropriate RAM, a suitable processor, a suitable graphics card, or a suitable logic unit, so that the respective method steps can be executed efficiently.

[0074] The computer program products are stored, for example, on a computer-readable storage medium or on a network or server, from where they can be loaded into the processor of the respective computing unit, which can be directly connected to the computing unit or formed as part of the computing unit. Furthermore, control information of the computer program products can be stored on a computer-readable storage medium. The control information of the computer-readable storage medium can be designed such that it carries out a method according to the invention when the data carrier is used in a computing unit. Examples of computer-readable storage media are a DVD, a magnetic tape, or a USB stick on which electronically readable control information, in particular software, is stored.If this control information is read from the data carrier and stored in a computing unit, all inventive embodiments / aspects of the methods described above can be implemented. Thus, the invention can also be based on said computer-readable medium and / or said computer-readable storage medium. The advantages of the proposed computer program products or the associated computer-readable media essentially correspond to the advantages of the proposed methods.

[0075] Further features and advantages of the invention will become apparent from the following explanations of exemplary embodiments based on schematic drawings. Modifications mentioned in this context can be combined with one another to form new embodiments. In different figures, the same reference numerals are used for the same features.

[0076] They show: Fig. 1 a schematic representation of an embodiment of a system for providing medical information, Fig. 2 a flowchart of a method for providing medical information according to an embodiment, Fig. 3 Dependencies of different data sets and information according to one embodiment, Fig. 4 Dependencies of different data sets and information according to one embodiment, Fig. 5 a flowchart of a method for providing medical information according to an embodiment, Fig. 6 a flowchart of a method for providing a trained function according to one embodiment, Fig. 7 a graphical user interface for a method for providing medical information according to one embodiment, Fig. 8a graphical user interface for a method for providing medical information according to one embodiment, and Fig. 9 a graphical user interface for a method for providing medical information according to one embodiment.

[0077] In Figure 1A system 1 for providing medical information MEDI based on a patient's medical data set PDS and on medical comparison data sets VDS is shown according to one embodiment. The system 1 has a user interface 10, a computing unit 20, an interface 30, and a storage unit 50. The computing unit 20 is fundamentally designed to generate and provide the medical information MEDI based on the medical data set PDS and the comparison data sets VDS. The comparison data sets VDS can be provided to the computing unit 20 via the interface 30 from the storage unit 50. The medical data set PDS can be provided to the computing unit 20 via the interface 30 and / or via the user interface 10 (e.g., by "uploading" the medical data set PDS).Alternatively, the user interface 10 can provide the computing unit 20 with data information about the medical data set PDS, and the computing unit can load the medical data set PDS from a corresponding database (not shown) using the data information. According to some embodiments, this database can be part of the storage unit 50.

[0078] The storage unit 50 can be configured as a centralized or decentralized database. The storage unit 50 can, in particular, be part of a server system. The storage unit 50 can, in particular, be part of a medical information system, such as a hospital information system (HIS), a PACS system (PACS stands for picture archiving and communication system), and / or a laboratory information system (LIS). The storage unit 50 can also be configured as a so-called cloud storage system. The storage unit 50 is designed to store a number of comparison data sets (VDS). Comparison data sets (VDS) relate to comparison cases with known or verified medical findings MD1, MD2, MD3. The comparison data sets (VDS) can have the same format as the medical data set (PDS) of the patient to be diagnosed.

[0079] The medical dataset (PDS) represents the patient data of the patient to be diagnosed. The comparison datasets (VDS) are patient data of real and / or virtual / synthetic comparison patients (who are different from the patient to be diagnosed). The medical dataset (PDS) and the comparison datasets (VDS) may contain medical image data and / or other medical data that does not contain image information. Image data in this context may refer to medical image data with two or three spatial dimensions. Furthermore, the image data may also contain a temporal dimension. The image data may, for example, have been generated using a medical imaging modality, such as an X-ray, computed tomography, magnetic resonance imaging, positron emission tomography, or angiography device, or other devices. Such image data may also be referred to as radiology image data.

[0080] Furthermore, the medical data set (PDS) and / or the comparison data sets (VDS) can also include histopathology image data, each of which shows one or more histopathology images. Histopathology image data is image data based on a patient's tissue sample. Tissue sections are prepared from the tissue sample and stained with a histological stain. The prepared tissue sections are then digitized to obtain the histopathology image data. Specialized scanners, so-called slide scanners, can be used for this purpose. The image acquired in this way is also referred to as a "whole slide image." The image data acquired in this way is typically two-dimensional pixel data.

[0081] The image data contained in the medical data set (PDS) or the comparison data sets (VDS) can, for example, be formatted according to the DICOM format. DICOM (=Digital Imaging and Communications in Medicine) is an open standard for the communication and administration of medical image data and related data.

[0082] In addition to image data, the medical data set PDS and the comparison data sets VDS can also include non-image data. Non-image data can, for example, be examination results that are not based on medical imaging. This can include laboratory data, vital data, spirometry data, or neurological examination protocols. In addition, non-image data can include text data sets, such as structured and unstructured medical reports. Non-image data can also be patient-related data. This can, for example, include demographic information about the patient, such as their age, gender, or body weight. The non-image data can be embedded in the image data, for example, as metadata. Alternatively or additionally, the non-image data can also be stored in an electronic medical record (EMR) of the patient, i.e.separate from the image data. Such electronic medical records can be archived, for example, in the storage device 50 or in a separate storage device with which the computing unit 20 can be connected via the interface 30.

[0083] The user interface 10 can have a display unit 11 and an input unit 12. The user interface 10 can be designed as a portable computer system, such as a smartphone, tablet computer, or laptop. Furthermore, the user interface 10 can be designed as a desktop PC. The input unit 12 can be integrated into the display unit 11, for example in the form of a touch-sensitive screen. Alternatively or additionally, the input unit 12 can have a keyboard or a computer mouse and / or a digital pen. The display unit 11 is designed to display, in particular graphically, individual or multiple elements from the medical data set PDS or the comparison data sets VDS, the determined medical information MEDI, and steps relevant for providing the medical information MEDI.The user interface 10 is further configured to receive input from the user regarding medical information MEDI relevant for a diagnosis. The user can be a physician.

[0084] The user interface 10 has one or more processors 13 configured to execute software for controlling the display unit 11 and the input unit 12 to provide a graphical user interface (GUI) that enables the user to select a patient or the associated medical data set (PDS) for diagnosis, to enter user inputs (NE) for providing the medical information (MEDI) into the system 1, and to review the retrieved medical information (MEDI). The user can activate the software, for example, via the user interface 10, for example, by downloading it from an app store and / or executing it locally. According to further embodiments, the software can also be a client-server computer program in the form of a web application that runs in a browser.

[0085] The interface 30 can have one or more individual data interfaces that ensure the data exchange between the components 10, 20, 50 of the system 1. The one or more data interfaces can be part of the user interface 10, the computing unit 20 and / or the memory unit 50. The one or more data interfaces can have a hardware and / or software interface, e.g., a PCI bus, a USB interface, a FireWire interface, a ZigBee or a Bluetooth interface. The one or more data interfaces can have an interface of a communications network, wherein the communications network can have a local area network (LAN), for example, an intranet or a wide area network (WAN). Accordingly, the one or more data interfaces can have a LAN interface and / or a wireless LAN interface (WLAN or Wi-Fi).

[0086] The computing unit 20 may include a processor. The processor may include a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an image processing processor, an integrated (digital or analog) circuit, or combinations of the aforementioned components, and other devices for providing the medical information MEDI according to embodiments of the invention. The computing unit 20 may be implemented as a single component or may include multiple components operating in parallel or serially. Alternatively, the computing unit 20 may include a real or virtual group of computers, such as a cluster or a cloud. Such a system may be called a server system. Depending on the embodiment, the computing unit 20 may be configured as a local server or as a cloud server.Furthermore, the computing unit 20 can have a working memory, such as a RAM, for temporarily storing, for example, the medical data set PDS or the comparison data sets VDS. Alternatively, such a working memory can also be arranged in the user interface 10. The computing unit 20 is configured, for example, by computer-readable instructions, by design and / or hardware, such that it can execute one or more method steps according to embodiments of the present invention. In particular, the computing unit 20 can be configured to execute one or more trained functions described further below.

[0087] The computing unit 20 may comprise subunits or modules 21-24 which are designed to provide a user with medical information MEDI within the framework of a continuous human-machine interaction and thus to support him in the diagnosis.

[0088] Module 21 is designed to provide reference data sets RDS-1, RDS-2, ..., RDS-n. Module 21 can be referred to as a similarity analysis module. For this purpose, module 21 can be designed to access the storage unit 50 and select one or more reference data sets RDS-1, RDS-2, ..., RDS-n from the comparison data sets VDS. The reference data sets RDS-1, RDS-2, ..., RDS-n are characterized by the fact that they exhibit a certain similarity to the medical data set PDS. For this purpose, module 21 can be designed, for example, to extract a data descriptor from the medical data set PDS and compare it with corresponding data descriptors from the comparison data sets VDS. A data descriptor can be understood as a data vector or feature vector in which relevant features for comparisons of different medical data sets are aggregated.Such relevant features can be extracted from both the image data and the non-image data of the medical dataset PDS, or can include, for example, demographic information about the respective patient (such as age or gender), laboratory values (such as the patient's PSA value), and / or vital data of the patient. Features extracted from image data can include, for example, image information such as patterns, color information, intensity values, etc. The data descriptor generated from the medical dataset PDS can be compared by module 21 with corresponding data descriptors of the comparison datasets VDS in order to identify cases among the comparison datasets VDS that exhibit a certain similarity to the medical dataset PDS of the patient to be diagnosed. For this purpose, module 21 can determine a similarity measure AE-1, AE-2, ... for each of the considered comparison datasets VDS., AE-n, which indicates (or quantifies) a similarity of the medical data set PDS with the respective comparison data sets VDS. Comparison data sets VDS with a certain similarity measure AE-1, AE-2, ..., AE-n (e.g. a similarity measure AE-1, AE-2, ..., AE-n above a certain threshold value) are identified as reference data sets RDS-1, RDS-2, ..., RDS-n. To carry out the similarity analysis, the module 21 can be designed to execute a first data processing algorithm ALG-1. The first data processing algorithm ALG-1 is then designed to identify one or more reference data sets RDS-1, RDS-2, ..., RDS-n based on the medical data set PDS and the comparison data sets VDS and to provide corresponding similarity measures AE-1, AE-2, ..., AE-n. The first data processing algorithm ALG-1 can have one or more trained functions.

[0089] Module 22 is designed to determine a set of medical attributes ATT-1, ATT-2, ..., ATT-m based on the similarity analysis from module 21. Module 22 can be referred to as a classifier module. For this purpose, one or more medical attributes ATT-1, ATT-2, ..., ATT-m can be assigned to each of the previously known medical findings MD1, MD2, MD3 associated with the reference data sets RDS-1, RDS-2, ..., RDS-n. For this purpose, module 22 can be designed to use a fixed assignment or assignment rule that assigns one or more different medical attributes ATT-1, ATT-2, ..., ATT-m to each possible medical finding MD1, MD2, MD3. The medical attributes ATT-1, ATT-2, ..., ATT-m determined for the medical findings MD1, MD2, MD3 associated with the reference data sets RDS-1, RDS-2, ..., RDS-n form the set of medical attributes ATT-1, ATT-2, ..., ATT-m, which is subsequently displayed to the user. A medical attribute ATT-1, ATT-2, ..., ATT-m can refer to a characteristic, a property, an accompanying circumstance or a (necessary) condition of the respective finding to which it is assigned. A medical attribute ATT-1, ATT-2, ..., ATT-m can apply to one or more of the previously known medical findings MD1, MD2, MD3. In other words, a medical attribute ATT-1, ATT-2, ..., ATT-m can thus apply, for example, to a first subset of the medical findings MD1, MD2, MD3 assigned to the reference data sets RDS-1, RDS-2, ..., RDS-n, while it does not apply to a second subset of the medical findings MD1, MD2, MD3 assigned to the reference data sets RDS-1, RDS-2, ..., RDS-n or even excludes them. Can the user of this medical attribute ATT-1, ATT-2, ..., ATT-m for the patient to be diagnosed, the previously known medical findings MD1, MD2, MD3 of the first subset become more relevant for the present case, while the previously known medical findings MD1, MD2, MD3 of the second subset become less relevant or can even be excluded. Conversely, the medical findings MD1, MD2, MD3 of the first subset can be excluded if the user can exclude the medical attributes ATT-1, ATT-2, ..., ATT-m for the patient to be diagnosed. A medical attribute ATT-1, ATT-2, ..., ATT-m can, for example, be a semantic identifier, such as a symbol, a word, or a (preferably short) phrase. The user can exclude or confirm the attributes ATT-1, ATT-2, ..., ATT-m by a corresponding user input NE (via module 24), which can quickly further narrow down the medical findings indicated on the basis of the similarity analysis.

[0090] Module 23 is designed to determine or adapt medical information I-MEDI / MEDI based on the available input variables, which supports the user in making a medical diagnosis. Module 23 can therefore be referred to as an information generation module. Initially, the available input variables are the similarity measures AE-1, AE-2, ..., AE-n of the reference data sets RDS-1, RDS-2, ..., RDS-n and the previously known medical findings MD1, MD2, MD3 assigned to the reference data sets RDS-1, RDS-2, ..., RDS-n. Through continued user-system interactions, the medical attributes ATT-1, ATT-2, ..., ATT-m and the corresponding user inputs NE are added as additional input variables. A similarity analysis, initially based only on the similarity measures AE-1, AE-2, ..., AE-n and the previously known medical findings MD1, MD2, MD3 is hereinafter also referred to as incident medical information I-MEDI. By assigning the medical attributes ATT-1, ATT-2, ..., ATT-m and taking into account the corresponding user inputs NE, this incident medical information I-MEDI can be adapted and ultimately made available as medical information MEDI. In this context, the module 23 can, for example, be designed to create a ranking of the previously known medical findings MD1, MD2, MD3 based on the similarity measures AE-1, AE-2, ..., AE-n and to output this as medical information MEDI. The module 23 can furthermore be designed to adapt the ranking as a result of the user input NE if it is adjusted due to the application or non-application of some medical attributes ATT-1, ATT-2, ..., ATT-m.The module 23 can be configured to repeatedly adapt the medical information MEDI when a new user input NE is present. To determine the medical information I-MEDI / MEDI, the module 23 can be configured to implement a second data processing algorithm ALG-2. The second data processing algorithm ALG-2 is configured to generate a piece of medical information I-MEDI / MEDI based on the medical findings MD1, MD2, MD3 as well as the similarity measures AE-1, AE-2, ..., AE-n and / or the user input NE. In particular, the second data processing algorithm ALG-2 can be implemented as an electronic classifier. Furthermore, the second data processing algorithm can have one or more trained functions.Furthermore, the module 23 can be designed to automatically analyze the medical data set PDS of the patient to be diagnosed to determine whether one or more of the medical attributes ATT-1, ATT-2, ..., ATT-m applies or does not apply.

[0091] Module 24 can be understood as a user interaction and / or visualization module. Module 24 is designed to display the medical information MEDI and / or the incident medical information I-MEDI as well as the determined medical attributes ATT-1, ATT-2, ..., ATT-m to the user. Module 24 is further designed to receive a user input NE regarding the medical attributes ATT-1, ATT-2, ..., ATT-m and to forward it to module 23. For this purpose, module 24 can be designed to provide a graphical user interface GUI in which the incident medical information I-MEDI or the medical information MEDI as well as the medical attributes ATT-1, ATT-2, ..., ATT-m and, if applicable, further information / data objects can be displayed and via which a user input NE can be entered. In addition, module 24 can be designed to archive the medical information MEDI (e.g.in the storage unit 50 or any other database) or to another module or another software for further processing. In particular, this can provide feedback to a trained function ALG-1, if used in the similarity analysis, with which the trained function ALG-1 can be further optimized.

[0092] The division of the computing unit 20 into modules 21-24 serves merely to simplify the explanation of the functionality of the computing unit 20 and is not to be understood as limiting. The modules 21-24 or their functions can also be combined into a single element. The modules 21-24 can in particular also be understood as computer program products or computer program segments, which, when executed in the computing unit 20, implement one or more of the method steps described below.

[0093] The computing unit 20 and the processor 13 together can form the controller 40. It should be noted that the illustrated layout of the controller 40, i.e., the described division into the computing unit 20 and the processor 13, is also to be understood only as an example. Thus, the computing unit 20 can be fully integrated into the processor 13 and vice versa. In particular, the method steps can run entirely on the processor 13 of the user interface 10 by executing a corresponding computer program product (e.g., software installed on the user interface), which then interacts directly, for example, with the memory unit 50 via the interface 30. In other words, the computing unit 20 would then be identical to the processor 13.

[0094] As already mentioned, according to some embodiments, the computing unit 20 can alternatively be considered a server system, such as a local server or a cloud server. In such an embodiment, the user interface 10 can be referred to as a "frontend" or "client," while the computing unit 20 can then be considered a "backend." Communication between the user interface 10 and the computing unit 20 can then be implemented, for example, based on an https protocol. The computing power in such systems can be divided between the client and the server. In a "thin client" system, the server has the majority of the computing power, while in a "thick client" system, the client provides more computing power. The same applies to the data (here: in particular, the medical data set PDS and the comparison data sets VDS).While in a "thin client" system the data mostly remains on the server and only the results are transmitted to the client, in a "thick client" system data is also transmitted to the client.

[0095] According to further embodiments, the described functionality can also be provided as a so-called cloud service or web service. The correspondingly configured computing unit is then configured as a cloud or web platform. The data to be analyzed, i.e., the medical data set (PDS), can then be uploaded to this platform via a suitable web interface.

[0096] In Figure 2A schematic flow diagram of a method for providing medical information MEDI is shown. The sequence of the process steps is not limited by the sequence shown or by the selected numbering. Thus, the order of the steps can be reversed if necessary, and individual steps can be omitted. Furthermore, one or more steps, in particular a sequence of steps, and optionally the entire process, can be repeated. Figure 3 and 4 Corresponding diagrams are shown, which show the Figure 2 The data streams associated with the methods shown in the Figures 3 and 4The reference data sets RDS-1, RDS-2, ..., RDS-n, medical attributes ATT-1, ATT-2, ..., ATT-m, and medical findings presented are only examples. In particular, depending on the case, different numbers of reference data sets RDS-1, RDS-2, ..., RDS-n, medical attributes ATT-1, ATT-2, ..., ATT-m, and medical findings MD1, MD2, MD3, and different assignments will result.

[0097] The Figure 2In some embodiments, the method presented aims to offer a selection of possible diagnoses / medical findings based on an automated analysis of the data available for a patient and to narrow them down through ongoing human-machine interaction. The available data is provided, on the one hand, by the medical data set PDS of the patient to be diagnosed and, on the other hand, by comparison data sets VDS. As already explained, each of the comparison data sets VDS relates to medical data sets for comparison patients (different from the patient to be diagnosed). The comparison data sets VDS are characterized in that for each comparison data set VDS, at least one previously known medical diagnosis MD1, MD2, MD3 is known, which was made for the associated comparison patient in the past.

[0098] In a first step S10, the medical data set PDS of the patient to be diagnosed is provided. This may involve a manual selection of the respective case by a user via user interface 10. Furthermore, this may involve loading the medical data set PDS from storage unit 50 or another data storage device, which may, for example, be part of a medical information system. Furthermore, step S10 may involve receiving the medical data set PDS by computing unit 40.

[0099] According to one idea of the present invention, the medical data set PDS is automatically compared by the computing unit 40 with the comparison data sets VDS in order to identify data sets among the comparison data sets VDS that exhibit a certain similarity to the medical data set PDS. These similar data sets are also called reference data sets RDS-1, RDS-2, ..., RDS-n. Assuming that similar medical data sets indicate similar medical findings MD1, MD2, MD3, possible medical findings MD1, MD2, MD3 for the medical data set PDS can be deduced from the previously known medical findings MD1, MD2, MD3 of the reference data sets RDS-1, RDS-2, ..., RDS-n.

[0100] In step S20, one or more reference data sets RDS-1, RDS-2, ..., RDS-n are determined from the comparison data sets VDS, which exhibit a certain degree of similarity to the medical data set PDS. Step S20 is based on a similarity analysis in which similarities between the medical data set PDS and the comparison data sets VDS are identified or quantified. The similarities can be expressed by similarity measures AE-1, AE-2, ..., AE-n. In particular, for at least a subset of the comparison data sets VDS, a similarity measure AE-1, AE-2, ..., AE-n can be determined, each of which expresses a similarity between the respective comparison data set VDS and the medical data set PDS. All comparison data sets VDS with a certain degree of similarity to the medical data set PDS can be used as reference data sets RDS-1, RDS-2, ..., RDS-n can be identified that is potentially relevant for the diagnosis of the medical data set PDS. For example, all those comparison data sets VDS whose similarity measure lies above a predetermined or predeterminable threshold can be identified as reference data sets RDS-1, RDS-2, ..., RDS-n. Alternatively, the comparison data sets VDS with the comparatively highest similarity measures can be identified as reference data sets RDS-1, RDS-2, ..., RDS-n. Step S20 thus carries out, as in . Figure 4 shown, to one or more reference data sets RDS-1, RDS-2, ..., RDS-n. Each reference data set RDS-1, RDS-2, ..., RDS-n has a corresponding similarity measure AE-1, AE-2, ..., AE-n. According to embodiments of the invention, the similarity analysis in step S20 is performed by the first data processing algorithm ALG-1. A more detailed description of the determination of the similarity measures is also provided with reference to Figure 5 given.

[0101] Based on the similarity analysis, the medical findings MD1, MD2, MD3 associated with the reference data sets RDS-1, RDS-2, ..., RDS-n are also relevant for the patient to be diagnosed. As described in Figure 3 As shown, each reference data set RDS-1, RDS-2, ..., RDS-n is associated with a medical finding MD1, MD2, MD3. Different reference data sets RDS-1, RDS-2, ..., RDS-n can be associated with identical medical findings MD1, MD2, MD3. According to some examples, a reference data set RDS-1, RDS-2, ..., RDS-n can also be associated with more than one medical finding (in Figure 3(not shown). The similarity measures AE-1, AE-2, ..., AE-n provide an indication of how likely a medical finding MD1, MD2, MD3 from a reference data set RDS-1, RDS-2, ..., RDS-n is also applicable to the medical data set PDS. If a medically certain finding MD1, MD2, MD3 can be associated with consistently high similarity measures AE-1, AE-2, ..., AE-n, it is more relevant than a medical finding MD1, MD2, MD3 with lower similarity measures AE-1, AE-2, ..., AE-n.

[0102] In step S30, this relationship is used to derive a first or incident medical information item I-MEDI, which provides the user with information about possible medical findings MD1, MD2, MD3 for the patient being diagnosed and, optionally, their potential relevance or pertinence. The relevance or pertinence can be derived from the similarity measures AE-1, AE-2, ..., AE-n, as explained above. For this purpose, the similarity measures AE-1, AE-2, ..., AE-n can be evaluated in step S30, for example, to obtain a confidence value for the respective medical finding MD1, MD2, MD3. This confidence value can be implemented as a "score" for the respective medical finding MD1, MD2, MD3. A confidence value can be implemented, for example, by calculating average similarity measures AE-1, AE-2, ..., AE-n or a possibly weighted sum of the similarity measures AE-1, AE-2, ..., AE-n.On this basis, the incident medical information I-MEDI can, for example, include a ranking of the medical findings MD1, MD2, MD3. The ranking would then indicate how accurate a respective medical finding MD1, MD2, MD3 is compared to other medical findings MD1, MD2, MD3 based on the similarity analysis. In . Figure 4In the example shown, in the incident medical information I-MEDI, the medical finding MD3 is ranked first and thus, based on the similarity analysis (the similarity measures AE-1, AE-2, ..., AE-n), is assessed as the most relevant for the patient to be diagnosed. Medical finding MD1 comes second, and medical finding MD2 comes third. Alternatively or additionally, the initial medical information I-MEDI can also contain the aforementioned confidence values. According to embodiments of the invention, the initial medical information I-MEDI can be provided by applying the second data processing algorithm ALG-2.

[0103] In an optional sub-step S31, the incident medical information I-MEDI can be displayed to the user via the user interface 10. This is preferably done in a form of presentation that is suitable for the user to perceive the relevance of a respective medical finding MD1, MD2, MD3 in comparison to the other medical findings MD1, MD2, MD3. As described below with reference to Figures 6 to 8 As will be described in more detail, you can, for example, work with the display of lists and / or numerical values and / or markings.

[0104] The medical findings MD1, MD2, MD3 found by the similarity analysis and their relative relevance assessment (i.e., the initial medical information I-MEDI) are based on the available data. It is generated automatically and does not yet contain any user input. Embodiments of the present invention provide for the initial medical information I-MEDI to be refined through ongoing and targeted human-machine interaction. This involves offering the user tags or medical attributes ATT-1, ATT-2, ..., ATT-m. The user can then decide, through a user input NE, whether and, if so, to what extent the medical attributes ATT-1, ATT-2, ..., ATT-m apply to the patient to be diagnosed. This assessment is usually much simpler than assessing whether a complete medical finding MD1, MD2, MD3 applies or not.The specification of incident medical information I-MEDI is therefore broken down into sub-problems that are easier for the user to handle. The medical attributes ATT-1, ATT-2, ..., ATT-m refer to higher-level properties or characteristics of the respective medical finding MD1, MD2, MD3. The medical attributes ATT-1, ATT-2, ..., ATT-m can refer to various categories related to medical findings. For example, medical attributes ATT-1, ATT-2, ..., ATT-m can include information on symptoms, clinical signs or contraindications, patient information, risk factors, differential diagnoses, etc. With regard to the clinical course, a medical attribute can, for example, indicate whether an acute, sub-acute, or chronic disease is present. Using a lung disease as an example, an attribute can indicate whether a clinical picture is unilateral or bilateral. An attribute ATT-1, ATT-2, ..., ATT-m of a medical finding MD1, MD2, MD3 can also be how frequently or rarely the underlying clinical picture occurs or whether an infectious disease is present. As already mentioned, the attributes ATT-1, ATT-2, ..., ATT-m can particularly be used as semantic identifiers. As in the . Figures 3 and 4 As shown, a medical finding can have several different medical attributes ATT-1, ATT-2, ..., ATT-m. Furthermore, a medical attribute ATT-1, ATT-2, ..., ATT-m can apply to several different medical findings MD1, MD2, MD3. Furthermore, different medical attributes ATT-1, ATT-2, ..., ATT-m can be mutually exclusive.

[0105] In step S40, the medical attributes ATT-1, ATT-2, ..., ATT-m relevant for further narrowing down the medical findings MD1, MD2, MD3 and specifying the incident medical information I-MEDI are determined. Relevant medical attributes ATT-1, ATT-2, ..., ATT-m are those medical attributes ATT-1, ATT-2, ..., ATT-m that are relevant for the medical findings MD1, MD2, MD3 determined by the similarity analysis. In an optional sub-step S41, an assignment or assignment rule can be provided that assigns one or more medical attributes ATT-1, ATT-2, ..., ATT-m to medical findings MD1, MD2, MD3. The assignment can, in particular, be known in advance and, for example, be created based on annotations by one or more experts and / or an evaluation of textbook knowledge or medical guidelines.In a further optional sub-step S42, the medical findings MD1, MD2, MD3 are each assigned these identifying medical attributes ATT-1, ATT-2, ..., ATT-m. As an alternative to this, an electronic (trained) classifier can be used which assigns one or more medical attributes ATT-1, ATT-2, ..., ATT-m to each medical finding MD1, MD2, MD3. As a further alternative, one or more medical attributes ATT-1, ATT-2, ..., ATT-m can be stored in the associated reference data sets RDS-1, RDS-2, ..., RDS-n and found by evaluating the reference data sets RDS-1, RDS-2, ..., RDS-n. As in . Figure 3As shown, the medical findings MD1, MD2, MD3 are obtained with their respective associated medical attributes ATT-1, ATT-2, ..., ATT-m. Optionally, these medical attributes ATT-1, ATT-2, ..., ATT-m can be displayed in step S30 or S90 together with the respective medical findings (see also Figures 6 to 8 ).

[0106] As in Figure 3As further shown, a set of medical attributes ATT-1, ATT-2, ..., ATT-m is created from the medical attributes ATT-1, ATT-2, ..., ATT-m of the medical findings MD1, MD2, MD3, which set is to be displayed to the user for assessment (sub-step S43). In doing so, for example, duplicate mentions of medical attributes ATT-1, ATT-2, ..., ATT-m can be eliminated. Optionally, medical attributes ATT-1, ATT-2, ..., ATT-m can also be filtered out. This can be particularly useful if a medical attribute only occurs in connection with a medical finding that has a low degree of similarity AE-1, AE-2, ..., AE-n, because this can mean that such a medical attribute ATT-1, ATT-2, ..., ATT-m will only have a low discriminatory effect for further narrowing down a medical finding MD1, MD2, MD3.

[0107] In a step S50, the medical attributes ATT-1, ATT-2, ..., ATT-m determined in step S40 are displayed to the user via the user interface 10. The purpose of this display is to provide the user with the determined medical attributes ATT-1, ATT-2, ..., ATT-m for verification and / or evaluation. Optionally, a display format can be selected in which the medical attributes ATT-1, ATT-2, ..., ATT-m are grouped by category and / or sorted by their discriminatory effect. The discriminatory effect can be determined, for example, based on the similarity measures AE-1, AE-2, ..., AE-n and / or simulation of a decision tree leading to one or more medical findings MD1, MD2, MD3. The verification or evaluation of the medical attributes ATT-1, ATT-2, ..., ATT-m by the user takes place by entering a corresponding user input NE.A user input NE can be directed to a confirmation and / or rejection or exclusion of individual medical attributes ATT-1, ATT-2, ..., ATT-m for the case to be diagnosed. In addition to such qualitative information, the user input NE can include a quantitative information for further differentiation, which contains a measure of how accurate individual medical attributes ATT-1, ATT-2, ..., ATT-m are for the case to be diagnosed (for example, such a measure can be selected from a range from 100% accurate to 0% accurate). Preferably, the display in step S50 is carried out via a graphical user interface (GUI), via which the user input NE can also be entered.

[0108] The user input NE is received in step S60. The user input NE can be provided by the user, for example, by pressing one or more buttons in a graphical user interface (GUI), as well as by text or voice input, or by other means by which data can be entered into the system 1.

[0109] In step S70, the user input NE is evaluated. In particular, in step S70, a medical information MEDI is created in which the user input NE is taken into account. This takes advantage of the fact that a verification or evaluation of a medical attribute ATT-1, ATT-2, ..., ATT-m by the user has a direct impact on the extent to which the associated medical finding MD1, MD2, MD3 applies to the present case. If the user input NE contains, for example, a selection of a specific medical attribute ATT-1, ATT-2, ..., ATT-m for the present case (in Figure 4shown example ATT2 and ATT3), this is to be interpreted as an indication that the medical findings MD1, MD2, MD3 associated with this medical attribute ATT-1, ATT-2, ..., ATT-m are more relevant for the case to be diagnosed than those that do not show (or even exclude) the medical attribute ATT-1, ATT-2, ..., ATT-m. The more medical attributes ATT-1, ATT-2, ..., ATT-m selected by the user apply to a medical finding MD1, MD2, MD3, the greater its potential relevance for the case to be diagnosed. Conversely, the exclusion of a medical attribute ATT-1, ATT-2, ..., ATT-m can lead to the exclusion of the medical findings MD1, MD2, MD3 that require this medical attribute ATT-1, ATT-2, ..., ATT-m.In other words, by evaluating the user input NE, a confidence value (e.g., in the form of a score) can be determined for each of the medical findings MD1, MD2, MD3 found by the similarity analysis. This confidence value indicates the relative relevance of the medical findings MD1, MD2, MD3 for the patient to be diagnosed. On this basis, similar to step S30, a piece of medical information MEDI can be determined that provides information about possible medical findings MD1, MD2, MD3 for the patient to be diagnosed and, optionally, their possible relevance or pertinence. The medical information MEDI can be based solely on the medical findings MD1, MD2, MD3 selected by the similarity analysis and the user input NE. In addition, the medical information MEDI can take into account the similarity measures AE-1, AE-2, ..., AE-n, thus combining the similarity analysis and the user input NE.The medical information MEDI can then be generated as incident medical information I-MEDI adapted based on the user input NE. According to embodiments of the invention, the medical information MEDI can be provided by applying the second data processing algorithm ALG-2.

[0110] The following step S80 is directed toward providing the medical information MEDI. This provision may include displaying the medical information MEDI, for example, via the user interface 10 (sub-step S81). Alternatively or additionally, the medical information MEDI may be provided for further processing (sub-step S82). Such further processing may include, for example, the (semi-)automated creation of a medical report, and in particular the automated completion of a report form. Furthermore, further processing may include archiving the medical information MEDI, preferably in association with the medical data set PDS. Furthermore, a further processing may include providing the medical information MEDI to the first data processing algorithm ALG-1.The underlying idea is that a user's narrowing down of the medical findings MD1, MD2, MD3 implicitly includes an evaluation of, for example, the similarity measures AE-1, AE-2, ..., AE-n found by the first data processing algorithm ALG-1. Thus, the medical information MEDI can be used to improve the first data processing algorithm ALG-1. In sub-step S82, it is preferable that the user must consent to or actively initiate the provision (e.g., by entering a value into the appropriately configured graphical user interface GUI).

[0111] Step S90 is a repetition step. If, for example, the user makes a further user input NE in response to the display in step S81, step S90 ensures that the medical information MEDI is adapted to the further user input NE (steps S60 and S70) and the provision is updated accordingly (step S80). The further user input NE can, for example, include the selection or deselection of further medical attributes ATT-1, ATT-2, ..., ATT-m.

[0112] Figure 5shows an exemplary embodiment for individual steps for determining reference data sets RDS-1, RDS-2, ..., RDS-n. The individual steps can be executed, in particular, within step S20. The order of the method steps is not limited by the illustrated sequence or the selected numbering. Thus, the order of the steps can be reversed if necessary, and individual steps can be omitted. Furthermore, one or more steps, in particular a sequence of steps up to all steps, can be executed repeatedly.

[0113] In a first step S21, a data descriptor is generated from the medical data set PDS. The data descriptor can comprise essential features of the medical data set PDS in the form of a feature vector. Since the medical data set PDS generally comprises image data and non-image data, the data descriptor can also be based on image features and non-image features. Image features can be extracted using image processing methods. These can include the identification, analysis, and / or measurement of objects, local and / or global structures, patterns, or textures contained in the image data of the medical data set PDS. The features can further comprise anatomical features and / or structures, such as the presence of an anatomical landmark or the size, texture, or density of an identified organ.Furthermore, the features can include parameters that characterize image values of the image data contained in the medical data set PDS. These can be, for example, parameters that describe a color, a grayscale, a contrast, or gradients of these variables. The features extracted from the non-image data can, for example, include metadata about the image data of the medical data set PDS. Furthermore, the features can relate to further context data of the patient to be diagnosed. These features can, for example, relate to demographic information about the patient, one or more pre-existing conditions, risk factors, existing diagnoses and findings, laboratory values, vital signs, etc. The data descriptor preferably has a plurality of features that together characterize the medical data set PDS. In some embodiments, step S21 is performed using the first data processing algorithm ALG-1.

[0114] In step S22 of Figure 5 The data descriptor is used to query the storage unit 50. For this purpose, the data descriptor can be compared with the corresponding data descriptors of the comparison data sets VDS. The corresponding data descriptors can be generated analogously to the data descriptor of the medical data set PDS and, in particular, can have the same structure as this. According to some embodiments, the corresponding data descriptors were already generated prior to the query in step S22 and stored in the storage unit 50 together with the comparison data sets VDS.

[0115] In the following step S23, a similarity metric is evaluated, with which a similarity between the data descriptor of the medical data set PDS and the corresponding data descriptors of the comparison data sets VDS can be quantified. All or only a portion of the comparison data sets VDS contained in the storage unit can be taken into account. In other words, in step S23, a similarity measure AE-1, AE-2, ..., AE-n can be obtained for each considered comparison data set VDS. According to some embodiments, the similarity metric can be a distance in the vector space between the feature vectors of the medical data set PDS and the comparison data sets VDS. For example, the distance can be given as a Euclidean distance. According to further examples, the similarity metric can be defined as the cosine similarity between the data descriptors.According to other examples, the similarity of individual features can be weighted individually.

[0116] In step S24, the similarity measures AE-1, AE-2, ..., AE-n determined in step S23 are used to select from the comparison data sets VDS those data sets that exhibit a certain similarity to the medical data set PDS. The selected data sets form the reference data sets RDS-1, RDS-2, ..., RDS-n used in steps S30 ff. According to exemplary embodiments, all comparison data sets VDS can be selected whose similarity measure is above a predetermined threshold.

[0117] The result of steps S21-S24 is therefore one or more reference data sets RDS-1, RDS-2, ..., RDS-n with associated similarity measures AE-1, AE-2, ..., AE-n, each of which indicates a similarity of the respective reference data set RDS-1, RDS-2, ..., RDS-n to the medical data set PDS. One or more - and in particular all of the steps S21-S24 - can be executed by the (appropriately designed) first data processing algorithm ALG-1. For this purpose, the first data processing algorithm ALG-1 can have one or more trained functions. Depending on the design, a trained function can execute more or fewer steps S21-S24. In particular, the extraction of the feature signatures can be carried out using a trained function, from which the similarity measures AE-1, AE-2, ..., AE-n are then calculated. Alternatively, a trained function can also be designed to directly specify the similarity measures AE-1, AE-2, ..., AE-n.In addition to trained functions, classical functions not based on artificial intelligence can of course also be used. One example would be so-called texture classification algorithms.

[0118] The trained functions used for the first data processing algorithm ALG-1 may, according to some embodiments, comprise a neural network. Neural networks may comprise a plurality of successive layers. Each layer comprises at least one, preferably multiple, nodes. Essentially, each node may perform a mathematical operation that maps one or more input values to an output value. The nodes of each layer may be connected to all or only a subset of nodes of a previous and / or subsequent layer. Two nodes are "connected" if their inputs and / or outputs are connected. The edges or connections are associated with a parameter, often referred to as a "weight" or "edge weight." Input values for the nodes of the respective first layer may be, for example, the medical dataset PDS and the comparison datasets VDS.The last layer is often referred to as the output layer. Depending on the design, the output values of the nodes of the output layer can be, for example, feature signatures (which would then have to be converted into similarity measures AE-1, AE-2, ..., AE-n) or simply similarity measures AE-1, AE-2, ..., AE-n. Between the input layer and the output layer are a number of hidden layers.

[0119] According to some embodiments, the trained functions may, in particular, comprise a convolutional neural network (CNN) or a deep convolutional neural network (DCN). Such trained functions then comprise one or more convolutional layers and, optionally, one or more deconvolutional layers. Furthermore, the trained function may comprise pooling layers, upsampling layers, and fully connected layers. Convolutional layers convolve the input and pass their result to the next layer by moving an image filter over the input.Convolutional layers can be particularly advantageous when searching for similar image regions, as in some embodiments. Pooling layers reduce the dimensions of the data by aggregating the outputs of groups of nodes in one layer to a single node in the next layer. Upsampling layers and deconvolution layers reverse the actions of the convolutional and pooling layers. Fully connected layers connect each node in previous layers to nodes in subsequent layers, essentially giving each node a "vote."

[0120] The second data processing algorithm ALG-2 can also include one or more trained functions. According to some embodiments, neural networks can also be described for this purpose, which are designed to output a medical indication I-MEDI / MEDI based on the available input data (i.e., the similarity measures AE-1, AE-2, ..., AE-n, the medical findings MD1, MD2, MD3, and / or the user inputs NE). In addition, the second data processing algorithm ALG-2 can have one or more trained classifiers, such as a k-nearest neighbor network or support vector machines (SVMs for short). Furthermore, the second data processing algorithm ALG-2 can of course also be based on a classical function that does not use artificial intelligence.

[0121] A trained function learns by adjusting parameters that determine the mapping from input to output data. In neural networks, these are, for example, the weights or weighting parameters (e.g., the edge weights) of individual layers and nodes. A trained function can be trained using supervised learning methods. For example, the method of backpropagation can be used here. During training, the trained function is applied to training input data to generate corresponding output values, the target values of which are known in the form of training output data.The difference between the output values and the training input data can be used to introduce a cost or loss functional as a measure of how well or poorly the trained function performs its assigned task. The goal of training is to find a (local) minimum of the cost functional by iteratively adjusting the parameters (e.g., the edge weights) of the trained function. This ultimately enables the trained function to deliver acceptable results over a (sufficiently) large cohort of training input data. This optimization problem can be solved using a stochastic gradient descent method or other approaches well known in the field.

[0122] For the first data processing algorithm ALG-1, if it has a trained function, training datasets would each contain medical training datasets and training comparison datasets, as well as, depending on the configuration of the first data processing algorithm ALG-1, associated verified feature signatures, verified reference datasets, and / or verified features. The verified reference datasets could be based on a user annotation made based on an analysis or diagnosis of the medical training dataset. The same applies to the verified similarity measures.

[0123] Training the first data processing algorithm ALG-1 according to some embodiments of the invention could then comprise applying the first data processing algorithm ALG-1 to the medical training data sets or training VDS to generate output values, as well as comparing the output values with the verified feature signatures, verified reference data sets, and / or verified similarity measures. Based on the comparison, one or more parameters of the first data processing algorithm ALG-1 can then be adjusted.

[0124] Furthermore, it may be provided that the first data processing algorithm ALG-1 is continuously trained using feedback from the medical information MEDI. A corresponding method is described in Figure 6The sequence of the process steps is not limited by the illustrated sequence or the chosen numbering. Thus, the order of the steps can be reversed if necessary, and individual steps can be omitted. Furthermore, one or more steps, in particular a sequence of steps up to and including all steps, can be repeated.

[0125] Step T10 is directed toward providing a medical data set PDS and several comparison data sets VDS. The medical data set PDS or the comparison data sets VDS can, in particular, have the form described above.

[0126] In step T20, the first data processing algorithm ALG-1 determines one or more reference data sets RDS-1, RDS-2, ..., RDS-n from the comparison data sets VDS. For example, as in connection with Figure 2executed, the reference data sets RDS-1, RDS-2, ..., RDS-n exhibit a certain similarity with the medical data set PDS. Furthermore, the first data processing algorithm ALG-1 outputs a similarity measure AE-1, AE-2, ..., AE-n for each reference data set RDS-1, RDS-2, ..., RDS-n, which indicates (quantifies) a similarity of the respective reference data set RDS-1, RDS-2, ..., RDS-n with the medical data set PDS. The first data processing algorithm ALG-1 can be used as described in connection with Figure 2 executed operate.

[0127] In step T30, based on the reference data sets RDS-1, RDS-2, ..., RDS-n determined in step T20 and the similarity measures AE-1, AE-2, ..., AE-n, a piece of medical information MEDI is generated through continued human-machine interaction. This can essentially be done as in Figure 2described, i.e. steps S30 to S80 can be processed repeatedly. If a sufficiently good result, i.e. acceptable medical information MEDI, is achieved, this can be used to further adapt the first data evaluation algorithm ALG-1. The medical information MEDI obtained in this way is optimized by the user inputs NE. Medical findings MD1, MD2, MD3, which have turned out to be (particularly) relevant for the patient to be diagnosed due to this optimization, suggest that reference data sets RDS-1, RDS-2, ..., RDS-n with such medical findings should show a greater similarity to the medical data set PDS than others. This can be exploited to check the first data evaluation algorithm ALG-1 and, if necessary, to further train it.

[0128] Accordingly, step T40 is directed at comparing the output of the first data processing algorithm ALG-1 with the medical information MEDI. If the medical information MEDI contains adjusted similarity measures, these can be directly compared with the similarity measures AE-1, AE-2, ..., AE-n determined by the first data processing algorithm ALG-1. Otherwise, possible adjustments to the similarity measures AE-1, AE-2, ..., AE-n can be inferred, for example, from the relative relevance of the medical findings MD1, MD2, MD3 in the medical information MEDI.

[0129] Finally, in step T50, the first data processing algorithm ALG-1 is adapted based on the comparison from step T40.

[0130] In the Figures 7 to 9exemplary embodiments of graphical user interfaces (GUI) are shown with which continuous human-machine interaction can be carried out according to some examples of the invention.

[0131] After a user has selected a case to be assessed, the first step is, as in connection with Figure 2described, a similarity analysis. The result of the similarity analysis is, on the one hand, the similar cases (in the form of one or more reference data sets RDS-1, RDS-2, ..., RDS-n) and the associated similarity measures AE-1, AE-2, ..., AE-n. On the other hand, one or more medical findings MD1, MD2, MD3 are obtained from the similarity analysis, which, based on the similarity analysis, are also relevant for the present case. In addition, the similarity analysis can provide information about how relevant the individual medical findings MD1, MD2, MD3 are assessed relative to one another. For example, a ranking of the medical findings MD1, MD2, MD3 for the present case can be derived from the similarity measures AE-1, AE-2, ..., AE-n. This information obtained from the similarity analysis can be collectively referred to as incident medical information I-MEDI.The incident medical information I-MEDI includes at least the medical findings MD1, MD2, MD3 determined in the similarity analysis and preferably also information about their (relative) relevance to the present case. Furthermore, the incident medical information I-MEDI can also include further information and data, such as data objects DO-A, DO-B, DO-C derived from the reference data sets RDS-1, RDS-2, ..., RDS-n, or other sources. These data objects can include image data and / or text data.

[0132] The incidental medical information I-MEDI can provide the user with an example in Figure 7 The graphical user interface GUI is shown. The display format used is preferably designed such that a ranking of the medical findings MD1, MD2, MD3 determined from the similarity analysis is visible to the user. In the Figures 7 to 9In the examples shown, this is achieved by displaying the medical findings MD1, MD2, MD3 in a list that is graded according to decreasing relative relevance. The medical finding that is initially most relevant for the case to be diagnosed based on the similarity analysis would be Figure 7 In the example shown, the medical finding is MD3. Alternatively or additionally, the relative relevance for each medical finding MD1, MD2, MD3 can be represented by visualizing a confidence value, which can be derived, for example, from the similarity measures AE-1, AE-2, ..., AE-n. This has the further effect of providing the user with an indication from System 1 about how clear the ranking is. As a further option, the ranking can be indicated additionally or alternatively by displaying the respective ranking number.

[0133] The medical findings MD1, MD2, and MD3 can specifically refer to diseases diagnosed for the reference data sets RDS-1, RDS-2, ..., RDS-n. Accordingly, the name of the respective disease can be specified when displaying I-MEDI in the graphical user interface (GUI). All medical findings MD1, MD2, and MD3 identified by the similarity analysis can be displayed, or only a subset—for example, the most relevant ones. The user can then display additional medical findings, for example, by "scrolling" through the list view.

[0134] In addition to the names of diseases (or any other identifiers of medical findings), the incident medical information I-MEDI can contain additional data objects DO-A, DO-B, DO-C and display them accordingly. The data objects DO-A, DO-B, DO-C can be generated, for example, from the reference data sets RDS-1, RDS-2, ..., RDS-n. In particular, the data objects DO-A, DO-B, DO-C generated from the reference data sets RDS-1, RDS-2, ..., RDS-n can include image data, which is then displayed as images in the graphical user interface (GUI) (see Figures 7 to 9). Furthermore, the data objects DO-A, DO-B, DO-C can be extracted from other sources, such as medical guidelines or electronic compendia or textbooks, such as eRef from Thieme or Radiopedia. In particular, such data objects DO-A, DO-B, DO-C generated from other sources can refer to the medical findings MD1, MD2, MD3 and, for example, provide more detailed information on the underlying disease. These data objects can be image data and / or, as described in the Figures 7 to 9 schematically represented, contain text data. The data objects DO-A, DO-B, DO-C are preferably presented for the I-MEDI display in such a way that the user can visually assign them to the respective corresponding medical findings MD1, MD2, MD3. As described in the Figures 7 to 9 As shown, the data objects DO-A, DO-B, DO-C can be grouped accordingly.

[0135] As detailed in connection with Figure 2As explained, the method according to the invention provides for assigning one or more medical attributes ATT-1, ATT-2, ..., ATT-m to the medical findings MD1, MD2, MD3. The respective medical attributes ATT-1, ATT-2, ..., ATT-m and their assignment to the respective medical finding can be contained in the I-MEDI. The medical attributes ATT-1, ATT-2, ..., ATT-m assigned to a medical finding MD1, MD2, MD3 can be displayed together with the respective medical finding, in particular in such a way that the assignment is clear to the user (cf. Figures 7 to 9 ).

[0136] As in the Figures 7 to 9As shown, the medical attributes ATT-1, ATT-2, ..., ATT-m can be represented as words that describe properties or characteristics of the respective medical finding MD1, MD2, MD3 and / or the reference data sets RDS-1, RDS-2, ..., RDS-n. Alternatively or additionally, characters, symbols, or word groups can also be used to represent the medical attributes ATT-1, ATT-2, ..., ATT-m.

[0137] Additionally, the medical attributes ATT-1, ATT-2, ..., ATT-m are displayed in a button 100. In button 100, the medical attributes ATT-1, ATT-2, ..., ATT-m can each be linked to a switch element 101, with which the user can make a user input NE related to the respective medical attribute ATT-1, ATT-2, ..., ATT-m. For easier reference for the user, the button 100 can be labeled with the word "Tags" or similar. In particular, the medical attributes ATT-1, ATT-2, ..., ATT-m linked to the medical findings MD1, MD2, MD3 are displayed in button 100. All medical attributes ATT-1, ATT-2, ..., ATT-m associated with the medical findings MD1, MD2, MD3, or only a selection of the associated medical attributes ATT-1, ATT-2, ..., ATT-m can be displayed. Furthermore, all intended or possible medical attributes ATT-1, ATT-2, ... can be displayed., ATT-m, for example, all medical attributes ATT-1, ATT-2, ..., ATT-m contained in the assignment rule can be displayed. In other words, the same medical attributes ATT-1, ATT-2, ..., ATT-m can always be displayed in button 100 of the graphical user interface (GUI), regardless of the case (i.e., in particular, regardless of the medical findings MD1, MD2, MD3 determined by the similarity analysis). The medical attributes ATT-1, ATT-2, ..., ATT-m can be displayed grouped by category for the sake of clarity. In the example shown, for example, the medical attributes "acute," "subacute," and "chronic" characterizing the course of the disease are grouped in the same way as the medical attributes "unilateral," "bilateral," and "focal" characterizing the site of onset of the disease. In addition, the medical attributes ATT-1, ATT-2, ..., ATT-m can be sorted by relevance ortheir potentially discriminatory effect.

[0138] The button 100 or the switching elements 101 are designed such that the user can enter a user input NE relating to individual medical attributes ATT-1, ATT-2, ..., ATT-m using the graphical user interface GUI. A user input NE can in particular comprise a verification of individual medical attributes ATT-1, ATT-2, ..., ATT-m. A verification can, for example, comprise a selection or confirmation of one or more medical attributes for the case to be diagnosed. Furthermore, a verification can comprise a deselection or exclusion of one or more medical attributes ATT-1, ATT-2, ..., ATT-m for the case to be diagnosed. For this purpose, the switching element 101 can be implemented, for example, as a button, where a single actuation (e.g., by mouse click) means a selection and a double actuation (e.g., by double-click) means a deselection of the respective medical attribute.Alternatively, an input option can be implemented using one or more radio buttons and / or other solutions. Furthermore, it can also be provided to provide the user with a more differentiated option for user input (NE), allowing them to, for example, adjust the extent to which individual medical attributes ATT-1, ATT-2, ..., ATT-m apply. Within the graphical user interface (GUI), this can be implemented, for example, by providing sliders.

[0139] In Figure 8 This example shows how the graphical user interface (GUI) can be adapted in response to a user input (NE). In this example, the user has confirmed the medical attributes "acute" and "infectious" using button 100. For better orientation, this input can be graphically identified (as in Figure 8(illustrated by check marks, for example). Accordingly, the medical attributes ATT-1, ATT-2, ..., ATT-m can also be marked for the respective medical findings MD1, MD2, MD3. The user input NE makes those medical findings MD1, MD2, MD3 more relevant that show one or more of the selected medical attributes ATT-1, ATT-2, ..., ATT-m. As a result of the user input NE, the medical information MEDI is therefore redefined or adjusted (see explanations on Figure 2 ) and the display in the graphical user interface (GUI) is updated. In the example shown, this results in medical findings MD1 and MD3 swapping places in the list because medical finding MD1 is now considered more relevant than medical finding MD3 due to the user input NE. Any displayed data objects DO-A, DO-B, and DO-C are updated if necessary.

[0140] In Figure 9This example shows how the graphical user interface (GUI) can be adapted in response to a user input (NE) if the user input (NE) also includes the exclusion of a medical attribute (ATT-1, ATT-2, ..., ATT-m). In the example shown, the user can "unilaterally" exclude the medical attribute for the case in question. This makes the medical finding (MD2) irrelevant for the case to be diagnosed, as it requires this medical attribute. The medical information (MEDI) is adapted accordingly. In the graphical user interface (GUI), this can be implemented, for example, by marking the medical finding (MD2) as "excluded."

[0141] While embodiments have been described in detail, particularly with reference to the figures, it should be noted that a multitude of modifications are possible. Furthermore, it should be noted that the exemplary embodiments are merely examples which are not intended to limit the scope of protection, application or structure in any way. Rather, the preceding description provides the person skilled in the art with a guide for implementing at least one embodiment, wherein various modifications, in particular alternative or additional features and / or modifications of the function and / or arrangement of the described components, can be made as desired by the person skilled in the art without deviating from the subject matter defined in the appended claims and its legal equivalents and / or leaving their scope of protection.

Claims

1. Computer-implemented method for the provision of an item of medical information (MEDI) with the steps: receiving (S10) a medical data record of a patient; determining (S20) one or more reference data records (RDS-1, RDs-2, ..., RDS-n) from a number of comparison data records (VDS) based on degrees of similarity (AE-1, AE-2, ..., AE-n), wherein a degree of similarity (AE-1, AE-2, ..., AE-n) is based on a similarity between the medical data record (PDS) and a comparison data record (VDS), and wherein each comparison data record (VDS) is associated with at least one already known medical finding (MD1, MD2, MD3); identifying (S40) one or more medical attributes (ATT-1, ATT-2, ..., ATT-m), which medical attributes (ATT-1, ATT-2, ..., ATT-m) comprise in each case an identifier of one or more of the already known medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDs-2), ...,RDS-n), wherein the step of identifying (S40) the one or more medical attributes comprises a determination of an attribute ranking based on the degrees of similarity (AE-1, AE-2, ..., AE-n) determined for the respective reference data records (RDS-1, RDS-2, ..., RDS-n) and / or based on a discriminatory effect of the medical attributes (ATT-1, ATT-2, ..., ATT-m) with respect to the relevance of the medical findings (MD1, MD2, MD3) for the patient to be diagnosed and / or based on one or more different attribute categories; displaying (S50) the identified medical attributes (ATT-1, ATT-2, ..., ATT-m) by way of a user interface (10), wherein the display (S50) of the identified medical attributes (ATT-1, ATT-2, ..., ATT-m) takes place in a display format (GUI) which is configured so that the user can perceive the attribute ranking; receiving (S60) a user input (NE) of a user relating to the displayed medical attributes (ATT-1, ATT-2, ..., ATT-m) by way of the user interface (10), wherein the user input (NE) has a verification of at least one part of the identified medical attributes (ATT-1, ATT-2, ..., ATT-m) comprising a confirmation and / or rejection of one or more of the identified attributes (ATT-1, ATT-2, ..., ATT-m); generating (S70) an item of medical information (MEDI) for the patient based on the medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDs-2, ..., RDS-n) and the user input (NE) comprising a determination of a ranking of at least one part of the already known medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDs-2, ..., RDS-n) based on the similarities (AE-1, AE-2, ..., AE-n) determined for the respective reference data records (RDS-1, RDs-2, ..., RDS-n) and based on the user input (NE); and providing (S80) the medical information (MEDI).

2. Method according to claim 1, in which the provision (S80) of the medical information (MEDI) comprises displaying (S81) the medical information (MEDI) by way of the user interface (10); wherein the display (S81) of the medical information (MEDI) takes in place in particular in a display format (GUI) which is embodied such that the user can perceive the ranking of the medical findings (MD1, MD2, MD3).

3. Method according to one of the preceding claims, in which the medical attributes (ATT-1, ATT-2, ..., ATT-m) comprise one or more of the following statements: an, in particular demographic, statement relating to a patient group associated with the clinical findings (MD1, MD2, MD3); a statement relating to one or more symptoms associated with the respective medical findings (MD1, MD2, MD3); a statement relating to one or more differential diagnoses associated with the respective medical findings (MD1, MD2, MD3); a statement relating to one or more clinical markers associated with the respective medical findings (MD1, MD2, MD3); and / or a statement relating to one or more clinical contra-markers associated with the respective medical findings (MD1, MD2, MD3).

4. Method according to one of the preceding claims, in which the step of identifying (S40) the one or more medical attributes (ATT-1, ATT-2, ..., ATT-m) comprises: providing an assignment (S41), which assigns one or more medical attributes (ATT-1, ATT-2, ..., ATT-m) to at least one already known medical finding (MD1, MD2, MD3); and assigning (S42) one or more different medical attributes (ATT-1, ATT-2, ..., ATT-m) to at least one of the medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDS-2, ..., RDS-n) using the assignment.

5. Method according to one of the preceding claims further with the step: generating (S30) an item of incident medical information (I-MEDI) based on the medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDS-2, ..., RDS-n), in particular based on the similarities (AE-1, AE-2, ..., AE-n) determined for the respective reference data records (RDS-1, RDS-2, ..., RDS-n), wherein the step of generating (S70) the medical information (MEDI) comprises an adjustment of the incident medical information (I-MEDI) based on the user input (NE).

6. Method according to one of the preceding claims, in which the medical data record (PDS) and the comparison data records (VDS) each comprise medical image data; and a degree of similarity (AE-1, AE-2, ..., AE-n) is based in each case on a similarity between the medical image data of the medical data record (PDS) and the medical image data of a comparison data record (VDS).

7. Method according to one of the preceding claims, in which the determination (S20) of the one or more reference data records (RDS-1, RDS-2, ..., RDS-n) comprises: extracting (S21) a data descriptor from the medical data record (PDS); receiving (S22) in each case a corresponding data descriptor for each of the comparison data records (VDS); determining (S23), for each comparison data record (VDS), a degree of similarity, wherein a degree of similarity is based in each case on a similarity between the data descriptor and a corresponding data descriptor; and determining (S24) the one or more reference data records (RDS-1, RDS-2, ..., RDS-n) based on the determined degrees of similarity.

8. Method according to one of the preceding claims, in which the determination (S20) of the one or more reference data records (RDS-1, RDS-2, ..., RDS-n) comprises an application of a trained function (ALG-1), which trained function (ALG-1) is embodied to determine a degree of similarity (AE-1, AE-2, ... AE-n) between medical data records (PDS, VDS).

9. Method according to claim 8, in which the step of providing (S80) comprises: providing (S81) the medical information (MEDI) to the trained function (ALG-1); further with the step: adjusting (T40, T50) the trained function (ALG-1) based on the medical information (MEDI).

10. System (1) for providing an item of medical information (MEDI) comprising: an interface (10, 30) and a controller (40) wherein the controller (40) is embodied: - to receive a medical data record (PDS) of a patient by way of the interface (10, 30); - to determine one or more reference data records (RDS-1, RDs-2, ..., RDS-n) from a number of comparison data records (VDS) based on degrees of similarity (AE-1, AE-2, .., AE-n), wherein a degree of similarity (AE-1, AE-2, ..., AE-n) is based on a similarity between the medical data record (PDS) and a comparison data record (VDS), and wherein each comparison data record (VDS) is associated with at least one already known medical finding (MD1, MD2, MD3); - to identify one or more medical attributes (ATT-1, ATT-2, ..., ATT-m), which medical attributes (ATT-1, ATT-2, ..., ATT-m) in each case comprise an identifier of one or more of the already known medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDs-2, ..., RDS-n) and to determine an attribute ranking based on the degrees of similarity (AE-1, AE-2, ..., AE-n) determined for the respective reference data records (RDS-1, RDS-2, ..., RDS-n) and / or based on a discriminating effect of the medical attributes (ATT-1, ATT-2, ..., ATT-m) in respect of the relevance of the medical findings (MD1, MD2, MD3) for the patient to be diagnosed and / or based on one or more different attribute categories; - to display the identified medical attributes (ATT-1, ATT-2, ..., ATT-m) by way of the interface (10, 30) in a display format which is configured so that the user can perceive the attribute ranking; - to receive a user input (NE) of a user relating to the indicated medical attributes (ATT-1, ATT-2, ..., ATT-m,) by way of the interface (10, 30), wherein the user input (NE) has a verification of at least one part of the identified medical attributes (ATT-1, ATT-2, ..., ATT-m) comprising a confirmation and / or rejection of one or more of the identified attributes (ATT-1, ATT-2, ..., ATT-m); - to generate an item of medical information (MEDI) for the patient based on the medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDS-2, ..., RDS-n) and the user input (NE) comprising a determination of a ranking of at least one part of the already known medical findings (MD1, MD2, MD3) associated with the reference data records (RDS-1, RDs-2, ..., RDS-n) based on the similarities (AE-1, AE-2, ..., AE-n) determined for the respective reference data records (RDS-1, RDs-2, ..., RDS-n) and based on the user input (NE); and - to provide the medical information (MEDI) by way of the interface (10, 30).

11. Computer program product which comprises a program and can be loaded directly into a memory of a programmable computing unit of a controller (40), having program means, in order to execute a method according to claims 1 to 9, when the program is executed in the controller (40).

12. Computer-readable storage medium, on which readable and executable program sections are stored, in order to execute all steps of the method according to one of claims 1 to 9, when the program sections are executed by the controller (40).